We examine the feasibility of a collaborative recommender system in the exercise domain targeted specifically at runners. By using a large dataset of over 600000 runners' finish times we explore the contrasts between casual and elite runners and hypothesise how a recommender system may be used to mitigate some of these diferences. We also briefly discuss some of the challenges faced by such a recommendation task and suggest how these challenges could be addressed.
• Information systems → Recommender systems; • Humancentered computing → Social recommendation;
Running is one of the most popular forms of exercise on the planet.
In 2015 over 17 million race finishers were recorded in the United
States alone 1. The proliferation of online resources promoting
running shows the growing popularity of the sport and is a persuasive
demonstration of the participants’ interest in improving their
performance, and the desire to improve the performance over time,
while avoiding injury, is a key motivation for us to address as a
recommender system problem [
Despite the popularity of running and the availability of training resources, it remains a dificult pastime. Completing a race is the end result of weeks or months of meticulous training and planning. Selecting the best balance of hard work, recovery, and diferent training types remains a concern that requires an in-depth knowledge of running and human performance. While runners will exhibit natural performance increases simply by starting to run and by improving their fitness, it is extremely dificult to further optimise performance without access to a coach or extensive time investment in training methodologies. The availability of coaches is limited and usually reserved for competitive athletes, rather than those simply trying to improve themselves and maximise their personal performance. There also exist many conflicting training methodologies and thus despite extensive research an unsuitable training programme may be selected by a runner.
In an attempt to counteract these problems we propose a novel collaborative recommender system. Our system is designed to recommend training plans and race strategies to a runner and thus alleviate the need for a coach and reduce the requirement for extensive research to devise a personalised training plan best suited to the runner. By mining the data on training plans of runners with similar histories we can suggest a training plan to the user that they can follow with little cognitive efort and that will allow them to make significant gains over a period of time.
In this paper, we show that elite runners progress faster than casual runners and we surmise that this is due to the fact they have access to coaches, research, and have a tendency to approach training in a more strategic manner. Imparting the expertise that elite runners have garnered to a casual runner through the use of a recommender system will lead to casual runners exhibiting increased rates of improvement, similar to that of the elite runner. We will also highlight the suitability of a collaborative approach to recommendation, outline the time scales involved in such recommendation and briefly suggest some approaches towards recommendation of material in this domain. While the work presented in this paper focuses on the marathon distance, the results are generalisable across diferent race distances, and as such running as a whole. 2
The use of machine learning for athletic performance is still in its early phases. Much of the work previously undertaken in this field has focused on the problem of prediction - the question of how fast a runner would run a race given a previous race they have run at a diferent distance.
One of the earliest works in this regard was undertaken by
Peter Riegel [
Despite its limitations Riegel’s formula is the most commonly used method for predicting race times today. It is the equation used by many well known websites, most notably RunnersWorld (www.runnersworld.com).
A further regression analysis was performed by David Cameron
[
A study [
Moving away from prediction into the field of recommendation,
Smyth [
As mentioned in the related work section the current methods
used for marathon prediction rely on data ranging from a single
race at each distance (i.e. the Riegel Model) to a few thousand self
reported race times. In contrast, we have built a dataset scraped
from various sources including race results tables and websites
allowing athletes to self declare race times. This dataset contains
over 600000 runners and their entire race histories. This is the first
dataset of this scale that has been collected and allows for the first
large scale data analysis and machine learning approach towards
making predictions for a marathon time. While we do not have
complete training histories for runners, which would give greater
resolution in a machine learning problem, we attempt to
approximate their training schedules by looking at the frequency of the
races of various non-marathon distances the athlete has run and
the improvements they make as they run them. Additional features,
such as age and gender, allow us to further distinguish between
runners in a way that has been neglected by many of the previous
prediction models mentioned in Section 2. These features,
particularly gender, have significant efects when it comes to distance
running [
plateau in performance. This natural performance improvement is a first indication of the potential benefits of a recommender system. Most runners do not have access to advanced training methods and lack the motivation or finances to employ a running coach, yet they still exhibit significant improvement from one race to the next. Finding suitable training plans requires a significant amount of research as there are many diferent methodologies runners have used in order to get results. A recommender system designed to assist runners could act in the role of a professional running coach and help personalise training plans based on data mining the successful training approaches of similar runners. The recommender system removes the time burden required to map out an adequate training plan and helps the runner to improve at the fastest rate. 3.3
After creating a basic user profile for each runner comprising
average race finish times at various non-marathon distances, we trained
two diferent models to predict marathon times. These models were
a simple K-Nearest Neighbours (KNN) model [
Figure 2 shows how these prediction errors change for athletes
with diferent race finish times. The Riegel model is indeed more
accurate than KNN or XGB for elite runners (finish times < 190mins),
but the vast majority of runners have slower finish times than this.
For these slower runners, the KNN and XGB models outperform
the Riegel model. This suggests that the use of user profiles and
computing similarity between them is a good approach for
describing how a runner is likely perform. This provides some justification
that a collaborative filtering system would provide an adequate
basis for building a recommender system for runners. The success of
Smyth’s work [
It could also be pointed out here that the simpler Riegel model out-performs both the KNN and XGB models at various points in the distribution. This is certainly the case for the quickest runners, which is not surprising as the Riegel model is based on world record times and we have little data at these points from which to build a similarity model. Similarly, for the very slowest runners, we also have a data sparsity issue. However, for the purposes of building a running recommender this is not a problem. The elite runners, with whom our model struggles, tend to be professional athletes or passionate runners. These runners tend to be well coached and well informed on their training and thus a recommender system is expected to be of limited value to them. Runners at the very slowest end of the spectrum tend to be one-of runners that are running purely for fun and they are also unlikely to see any value in using a recommender system. 3.4
We make the reasonable claim that elite runners either have access to coaching or are well-informed on training methods. As a result, they produce faster times but there are other phenomena that are a side efect of this optimised training. We define an elite runner as one who has run the qualifying standard for the Boston Marathon of 190 minutes, and then we compare the performance progression over time of elite runners and more casual runners. The Boston Marathon qualifying time was chosen as this is considered a goal time for many keen marathon runners and is a time that requires substantial training and efort to achieve. We also tell from Figure 2 that it corresponds roughly to the finish time at which the user based models begin outperforming the Riegel model which makes it a natural cut of.
In Figure 3 we show the point at which elite and casual runners ifrst run their overall Personal Best (PB) time since running their ifrst marathon. The elite runners clearly peak much earlier than casual runners, with nearly 20% of elites achieving a PB in the ifrst year after they start marathon running. In contrast, a higher fraction of casual runners achieve PB’s than elite runners, after a period of 5 years from their first marathon. In Figure 4 we show the age at which the personal best is achieved. Again, there is a strong diference between elite and casual runners, with elites much more likely to achieve the PB before the age of 40. Elite runners not only achieve their PB’s at a younger age but also at an earlier stage of their running career. This finding afirms the notion that the extra knowledge elite runners have over casual runners is a significant advantage when it comes to making performance gains. It is clear that a recommender system would be useful in this field - an automated personalised training plan generated by a collaborative recommender system would mitigate the need for a professional coach or extensive knowledge of running training for a casual runner. Such recommendations should lead to faster performance gains from casual runners and would see casual runners maximise their potential earlier.
It is important to note at this point that not even all elite runners maximise their potential quickly. Many elite runners will not run their personal best until up to five years after their first marathon. This demonstrates the potential time scale involved in such a recommender system with improvements not being apparent for months or even years after the first interaction with the system.
Such a recommendation system poses a unique challenge. How does a recommender system motivate a runner to keep using a system for a period of years, especially when the benefits of use may not be instant? Training for a marathon is dificult and a recommender may recommend a training session that, while clearly beneficial, may not lead to enjoyment or satisfaction for the user. The recommendation system must therefore keep a user engaged for long periods and convince them to make potentially unwanted decisions in order for them to see benefit.
An important factor in achieving this goal is to provide the user
with meaningful explanations. The ability of a system to make its
reasoning transparent contributes significantly to the users
acceptance of the recommendation [
In this paper we have examined the opportunity for a recommender system for runners. Elite runners improve at a faster rate through knowledge gained from coaching and research. This knowledge can be imparted to casual runners through the use of a recommender system, leading to greater levels of improvement for such runners. We have gathered an adequately large database of runners and race times and through this have shown that marathon times can be predicted by simple neighbourhood models. The better prediction accuracy of these models highlights the feasibility of a collaborative filtering approach to such recommendation. Lastly, we highlighted some of the issues with such a recommender system, namely the time scale involved. We suggested methods as to how such recommendation could be presented to runners in order to keep them suficiently motivated and allow adequate time for this recommendation to take efect.
In future we will implement such a recommender system. We expect that the results found will be generalisable to other endurance sports and as such we look to expand our research into activities such as swimming and cycling. The proliferation of wearable technology, such as heart rate monitors and GPS units, provides large quantities of data from training events and races. Such data will provide greater resolution for a machine learning approach and we will use such an approach to build a recommender system that can inform a user before, during, and after a training session or race.
This work is supported by Science Foundation Ireland through the Insight Centre for Data Analytics under grant number SFI/12/RC/2289